Working Paper Series∗ Department of Economics Alfred Lerner College of Business & Economics University of Delaware

Working Paper No. 2003-07

Predicting Interwar Business Cycles with the Interest Rate Yield Spread†

James L. Butkiewicz and Kim Lane Leong Long

September 2003

∗ †

http://www.be.udel.edu/economics/workingpaper.htm c 2003 by author(s). All rights reserved. °

Predicting Interwar Business Cycles with the Interest Rate Yield Spread

James L. Butkiewicz* University of Delaware and Kim Lane Leong Long University of Delaware

September 2003

JEL Classifications: E32, N12 Key Words: Business Cycles, Yield Spread, Great Depression

* Corresponding author: [email protected]. Abstract

Recent studies have demonstrated the ability of the interest rate yield spread to predict post-war business cycles. This same methodology is applied to the prediction of interwar business cycles in the United States and the United Kingdom. The spread predicts the early interwar cycles, although the lag in the United States is variable. The spread fails to provide a prediction of the 1937-1938 recession and the length of the depression in either country. Neither spread improves the recession forecast for the other country.

2

1. Introduction Recent research has demonstrated the usefulness of the slope of the yield curve for predicting real economic activity, both in the United States and in other countries. A partial list of these studies includes Davis and Fagan (1997), Estrella and Hardouvelis (1991), Estrella and Mishkin (1998), Hamilton and Kim (2002), Harvey (1988), Plosser and Rouwenhorst (1994), and Stock and Watson (1993). Estrella and Mishkin (1998) find that this variable dominates an extensive set of financial variables, including interest rates, stock prices and monetary aggregates, in predicting recessions during the post-war period. The yield spread, defined as a long-term interest rate minus a short-term interest rate of comparable risk, is normally positive, as the yield curve normally slopes upward. The normal positive slope is explained by the liquidity premium theory of the term structure of interest rates. As the curve flattens and/or inverts, the spread decreases and at times becomes negative. A flattening or inversion of the yield curve predicts that a recession is increasingly likely. This flattening of the spread is typically assumed to reflect a contractionary monetary policy. However, Estrella and Mishkin (1998) emphasize that the spread also contains information about expectations, specifically expectations about future monetary policy and about expected inflation. Hamilton and Kim (2002) note that in addition to monetary policy, the spread is affected by a timevarying term premium. The interwar period provides another sample to test the ability of the spread to predict business cycles. The testing procedure used here will be to use the spread to

predict recessions. Testing for discreet business cycle episodes enables use of monthly data for the interwar period, rather than the quarterly frequency that is typically used for post-war samples. Data are readily available to conduct tests for the United States and the United Kingdom. Investigation of the predictive ability of the spread for these two countries also facilitates tests of the international transmission of business cycles. That is, do the interest rate spreads in either country improve recession forecasts in the other country? This paper is organized as follows. Part 2 reviews the ability of the spread to predict post-war business cycles. Part 3 applies the same methodology to interwar cycles in the U.S. and U.K. In part 4, cross-country channels are explored. Part 5 summarizes and concludes. 2. The Spread and U.S. Post-war Cycles.1 As noted by Estrella and Mishkin (1998), most studies use the spread to predict quantitative economic variables, while they, and Stock and Watson (1993), predict recessions, a qualitative variable. Here we briefly review the usefulness of the term spread for forecasting recessions using monthly data for the post-war period.2 The model is a probit model, where the dependent variable equals one during recessions and zero otherwise. The contemporaneous or lagged interest rate spread with the best explanatory power is the independent variable. Estrella and Mishkin (1998) conclude that a parsimonious model’s predictive power often dominates that of a model containing more variables.

1 2

Estrella and Mishkin (1997) find the spread is also a useful predictor of post-war cycles in the U.K. Estrella and Mishkin (1998) use quarterly data and a slightly different sample period.

4

The probit model provides estimates of the probability of a recession: Pr( Yt = 1  Spreadt-i) = ( β0 + β1 Spreadt-i), i = 0 – 12.

(1)

The spread is the difference between the 10-Year Treasury Constant Maturity Rate and the 3-Month Treasury Bill Secondary Market Rate.3 Recession periods begin in the month of a cycle peak, and recoveries begin in the month of the cycle trough. Monthly cycle reference dates are from the NBER.4 The sample period is 1954.01 through 2003.06. This sample includes the most recent recession. The estimation procedure is to estimate the model for lags from zero to twelve months. The lag length with the best explanatory power is eight months.5 The results for this estimation are reported in Table 1.6 TABLE 1 Probit Estimation Sample: 1954.01 – 2003.06 Dependent Variable: Recession Variable Constant Spreadt-8

Coefficient -0.354 -0.837

t-statistic -3.906 -9.634

McFadden R2 0.286

More interesting are the recession forecasts provided by this simple model. The probabilities of recession are graphed in Figure 1.

3

Both series were obtained from the Federal Reserve Bank of St. Louis database, Fred II, at http://research.stlouisfed.org/fred2/. 4 The dates can be found at http://www.nber.org/cycles.html/. 5 Estrella and Mishkin (1998) obtain the best fit for a three quarter lag of the spread. Their psudo-R2 for the three quarter lag is 0.271, approximately the same value reported in Table 1. 6 Results for all estimates discussed in this paper are available from the corresponding author.

5

  $&% 

'()*,+- . /)#"0


 
 
    !#"


 
 


 



  

 



 





The results presented in Figure 1 demonstrate that the spread does provide good forecasts of the post-war recessions. The most notable error is the strong false positive signal in early 1967. 7 This reflects the “credit crunch” of late 1966. It should be noted that real GDP growth for the second quarter of 1967 was –0.3%.8, Smaller false positives occur in the late 1990s. The largest of these is comparable to the accurate recession signals of 1957 and 1960. The strongest recession signals are for recessions beginning in January 1980 and July 1981, followed by November 1973 and March 2001. The signals for the recessions beginning in August 1957, April 1960, and July 1990, while clearly

7

In their study of monetary trends, Friedman and Schwartz (1982) add a cycle to the NBER chronology with a peak in 1966 and a trough in 1967, for annual reference cycles. 8 The data are from the Bureau of Economic Analysis, Table S.1, at http://www.bea.doc.gov/bea/dn/nipaweb/TableViewFixed.asp?SelectedTable=1&FirstYear=2002&LastYe ar=2003&Freq=Qtr.

6

greater than preceding months, are relatively weak.9 Finally, the spread also provides accurate forecasts of recovery. Visual inspection of Figure 1 indicates that the forecast values fall to pre-recession levels at or near the cycle trough. 3. The Yield Spread and Interwar Business Cycles Does the spread retain its predictive power for interwar cycles? Data for both the United States and the United Kingdom are employed to investigate this question. The probit model used for the post-war sample is also used to estimate the predictive ability of the yield spread during the interwar period. For the United States, cycle reference dates are from the NBER, as above. No consistent interest rate series for short-term Treasury securities exists for the entire sample period of 1920.01 through 1938.12. Using private assets requires selecting assets of comparable risk. Thus, for the U.S., the spread between AAA bonds and the 4 to 6 month Prime Commercial Paper Rate is used. Both series are from Banking and Monetary Statistics, 1914-1941 (1943). For the United Kingdom, business cycle reference dates are from Burns and Mitchell (1946).10 The interest rate spread is the difference between the Yield of Consols and the Open Market Rates of Discount. Both interest rate series are from the NBER historical database.11 Probit models are estimated for both countries for the sample 1920.01 through 1938.12. In each case, the model is estimated for the contemporaneous spread as well as

9

The spread flattens, but does not become negative prior to these three recessions, while the yield spread does invert for all other recessions and during the 1966 “credit crunch.” 10 Morgenstern (1959) also uses these cycle reference dates. 11 These series are available at http://www.nber.org/databases/macrohistory/contents/uk.html.

7

one to twelve month lags. For the United States, the best fit is for a lag of seven months, very similar to the post-war result. For the United Kingdom, the best fit is for a fourmonth lag. The best estimate for the United States is reported in Table 2. As indicated by the McFadden R2, the fit for the interwar period is not as tight as for the post-war period. TABLE 2 Probit Estimation – United States Sample: 1920.01 – 1938.12 Dependent Variable: Recession Variable Constant Spreadt-7

Coefficient 0.123 -0.270

t-statistic 0.241 -4.282

McFadden R2 0.061

Figure 2 graphs the forecasted and actual recessions.

8

 $#%&  '
()
+* !-,).  /0//1!


 
 
    "!


 
 




 

 











 



For the interwar period, the timing between the yield spread and economic activity exhibits greater variability for the interwar period than for the post-war period. For first two recessions, the timing between the spread and the recession is approximately contemporaneous, while the lag is longer for other recessions.12 The strongest accurate recession signal is for the beginning of the Great Depression.13 The spread predicts the 1920-21 recession, but only contemporaneously, which reduces its value as a forecasting tool. However, the recession signal for the 1920-21 recession is ultimately stronger than that for the Great Depression. Romer (1990) argues that the extent of the initial downturn during the Great Depression was exacerbated by a decline in consumer durable goods 12

For estimates of the model for the sample period 1920.01 – 1928.12, the optimal forecast is obtained for a lag of one or two months. This is consistent with the Friedman and Schwartz (1963) finding of “long and variable lags.” 13 Hamilton (1987, p145) concludes, “…the depression was preceded by a dramatic shift toward a highly contractionary monetary policy.” As noted by Estrella and Mishkin (1998, p 50) and Hamilton and Kim (2002, p.346) the yield spread reflects, among several factors, the stance of monetary policy.

9

spending due income uncertainty resulting from the stock market crash.14 It is unlikely that the yield spread would reflect this real shock. An interesting feature of the forecast values is that the extent of the depression and the 1937-1938 recession are not forecast by the spread. In the post-war period, the spread provides a reasonably accurate forecast of both the beginning and the end of recessions. In the first three interwar cycles in the U.S., the spread reverses after or by the end of the cycle. After the 1929 contraction, the spread signals a recovery at least by the end of 1931. Moderate recession signals follow Britain’s departure from the gold standard in 1931 and the 1933 financial crisis. Neither is an especially strong signal.

Bernanke (1983) has demonstrated that the banking crises exerted contractionary pressure through the destruction of the normal channels of financial intermediation. This effect is an addition to the normal monetary forces reflected by the spread. Bordo, Erceg, and Evans (2000) and Cecchetti and Karras (1994) develop models explaining the output contraction during the Great Depression. Bordo, Erceg and Evans find that money shocks account for 50 to 70 percent of the decline in real output, but their model fails to capture the decline in output after early 1932. Their model predicts a stagnation of output beginning at that time, and they suggest that the worsening of the financial crises may be an additional explanatory factor not included in their model. Similarly, Cecchetti and Karras find that aggregate demand shocks can explain the decline in output through mid to late 1931, after which supply shocks better explain the output decline. They conclude

14

Hamilton (1987) also concludes that the initial downturn was more severe than can be attributed to contractionary monetary policy.

10

that the timing of the supply shocks is consistent with the destruction of credit channels by financial panics, as demonstrated by Bernanke. The results presented here for the spread are consistent with these findings. Hamilton (1987) argues that debt-deflation also contributed to the severity of the decline, in addition to the disruption of credit channels caused by the banking crises. Debt-deflation is another shock that is unlikely to be reflected by the yield spread. As deflation accelerated after 1930, this factor could also account for the failure of the spread to predict the length of the depression. There is almost no change in the forecast values for the 1937-38 recession; the spread does not provide any signal of this recession.15 A flight to quality (Mishkin, 1991), an increase in the term premium, and a liquidity trap are possible reasons why the spread fails to predict this recession. The overall results for the United Kingdom are similar. However, the lag is shorter, only four months. The estimate of the forecasting equation is reported in Table 3. TABLE 3 Probit Estimation – United Kingdom Sample: 1920.01 – 1938.12 Dependent Variable: Recession Variable Constant Spreadt-4

Coefficient 0.433 -0.483

t-statistic 0.3.480 -6.629

McFadden R2 0.155

15

Using the spread between long-term Treasury securities and 3-month bills similarly provides no signal of a recession at this time. Data for these series are from FRED II.

11

The actual and forecast recessions are displayed in Figure 3.

$#%&  '( !) +*-,.!/012 3 4!


 
 
    "!


 
 




 

 











 



The spread predicts the first three interwar recessions reasonably accurately. The signal for the end of the 1924-1926 recession is weak, but it is quickly followed by another recession. The prediction for the Great Depression is about as strong as for the 19201921 recession. As is the case for the U.S., the spread forecasts the beginning of the depression, but not the length. A recovery is signaled at least by 1931. Unlike the U.S., the U.K. did not experience financial crises that are a possible explanation for the continuation of depression, but it did experience deflation, making debt-deflation a likely contributing factor. The second rise in the spread, late in the depression, results from Britain’s departure from the gold standard in 1931.

12

As is the case for the United States, the spread provides no signal of the 19371938 recession. Even ahead of the United States, short-term interest rates in the United Kingdom had declined to very low levels (less than 1%) by mid 1932, and did not rise above 1% for the remainder of the sample. In the U.S., the commercial paper rate fell below 1% in mid 1934. For both countries, the spread fails to forecast the extent of the depression, and totally misses the downturns in 1937-1938. 4. International Transmission of Interwar Business Cycles Friedman and Schwartz (1963, 1982), Eichengreen (1992) and Eichengreen and Sachs (1985) have explained how the gold standard transmitted monetary shocks between nations during the interwar period. In particular, the failure of the United States and France to adhere to the rules of the game by expanding their money supplies in response to gold inflows increased the deflationary pressure transmitted to countries with balance of payments deficits. As also noted by Friedman and Schwartz (1963, 1982), the U.S. emerged as the dominant economy following World War I, and economic events in the United States were generally transmitted to the global economy. This suggests that, in particular, the U.S. spread may have predictive power for U.K. recessions. Influence from the U.K. to the U.S. is also possible. Were the interwar business cycles transmitted internationally? Certainly, three of the cycles are essentially coincident in the two countries, the 1920-21 recession, the Great Depression, and the 1937-38 recession. However, a system of fixed exchange rates works best with coordination of monetary policies, assuming the spread reflects, in part, monetary policy in the two countries. Thus, the period when policy and cycle

13

coordination is expected to be most synchronous is the gold standard period, 1925.04 through 1931.09.16 Simple correlations of the spreads for various time periods are reported in Table 4.

TABLE 4 Correlations of Interwar Spreads Sample Period 1919.01 - 1938.12 1925.04 - 1931.09 1925.04 - 1928.12 1919.01 – 1925.03, 1931.10 – 1938.12

Correlation 0.80 0.70 0.02 0.80

The correlation between the spreads is higher for the floating exchange rate period than it is for the fixed exchange rate period. The correlation is especially low for the 1925 –1928 period. To the extend that the spreads reflect monetary policy, policy coordination between the U.S. and the U.K. was the weakest during the first years following resumption of convertibility by the United Kingdom, just when the need for coordination was the greatest.17,18 The two yield spreads are graphed in Figure 4 for the full sample, 1919.01 – 1938.12.

16

Friedman and Schwartz (1963, 1982) note that even under the gold standard, domestic policies can deviate for periods of a few years. 17 It is well known that during this period, Britain generally kept interest rates high to defend an overvalued currency, while the United States enjoyed prosperity and essentially stable prices, and thus its monetary policy was more expansionary. The result is that policies and cycles followed divergent paths during this period, and exhibited much greater coordination during the remainder of the interwar period. 18 Another possibility is that the lead-lag timing between the United States and the United Kingdom changed following Britain’s return to gold. However, why this would be true for the 1924-1928 period, but not for 1929-1931 is not clear.

14

    "!#  %$'& ()* + ,! $-( . 


 







 












 
















The U.S. spread leads the U.K. spread in 1928, and the situation is reversed in 1931. The spreads move together in 1919 – 1920. The relationship between these three episodes is clearly evident, and well known. The spreads frequently move in divergent directions during the 1923 – 1928 period, and also after 1932. To further investigate the timing between the two spreads, Granger causality tests are run for both the levels and changes in the levels of the spreads for various lags.19 The results are very surprising. For all lags for both levels and changes in levels, the Granger causality tests reject the null hypothesis that the U.K. spread does not cause the U.S. spread. In no case is the null hypothesis that the U.S. spread does not cause the U.K.

19

In all cases, the test sample period is 1920.01-1938.12. For levels, eleven tests are run using from 2 through 12 lags. For changes, ten tests are run for 2 through 11 lags, as constrained by the data sample beginning 1919.01.

15

spread rejected. These unexpected results certainly are not consistent with accepted understanding of the transmission of interwar business cycles. Given the results of the Granger causality tests, the results of the probit estimates for either country are more surprising. For each country, a probit forecasting model is estimated using all combinations of the spread for each country, from zero to twelve lags. This is a total of 169 possible models for each country. In all possible cases for the United States, the U.K. spread is never statistically significant, in spite of the result that the U.K. spread Granger causes the U.S. spread. For the United Kingdom, for short lags of both spreads (e.g., the U.K. spread lagged four months and the U.S. spread lagged two months), both spreads are significant, but the U.S. spread has an incorrect positive coefficient.20 For long lags of the U.K. spread and various lags of the U.S. spread (e.g., the U.K. spread lagged eleven months and the U.S. spread from zero to twelve lags) the U.S. spread is significant with the correct sign, while the U.K. spread is insignificant, but with the correct sign. In none of these cases, however, is the explanatory power, as measured by the McFadden R2, or the predictions provided by the estimated modal superior to the model using only the U.K. spread lagged four months.21 Based on these results, interwar cycles are reasonably well predicted by each nation’s yield spread, and the addition of the other spread does not improve the estimates.

20

The U.S. spread is insignificant when the model is estimated omitting the 1925.04-1928.12 period. The anomalous result appears to be due to the divergent policies pursued during this time period. Another possibility, noted above, is that the lead-lag pattern between the two nations changed during this period. However, why this would be the case when the remaining cycles are essentially synchronous is unclear 21 As noted above, Estrella and Mishkin (1998) find that parsimonious specifications are generally superior to overfitted models.

16

Why this result obtains, when so much evidence supports a strong connection between nations for interwar cycles, is a topic for future research.

17

5. Conclusions The yield spread successfully predicts post-war recessions. For the interwar period, the yield spread is less successful forecasting cyclical turning points in the 1930s. For the United States, the spread provides an indication of recessions through the beginning of the Great Depression, but with a more variable lag than the post-war experience. The spread fails to predict the length of the depression or the 1937-1938 recession. Similarly, the spread provides forecasts for United Kingdom recessions with the same caveats. The lag is shorter for the U.K. than for the U.S., but less variable. As is the case for the United States, the predictive power of the spread breaks down for predicting the length of the depression and for the 1937-1938 recession. For countries with fixed exchange rates, monetary policies should be coordinated. The spreads are determined, in part, by monetary policies. While the spreads are highly correlated for the interwar sample, the correlation is the lowest in the years immediately following Britain’s return to the gold standard. However, for the 1920-1921 recession, the beginning of the Great Depression, and the period following Britain’s departure from the gold standard, the spreads move synchronously or display clear lead-lag patterns. In spite of the apparent correlation between the two spreads, neither improves recession forecasts in the other country In the final evaluation, the yield spreads do successfully predict most of the interwar business cycles. However, the lag is short in the U.K., and variable in the United States. The most glaring shortcomings are the inability of the spreads to predict the length of the depression and the 1937-38 recession in both countries.

18

References Bernanke, B. (1983), “Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression,” American Economic Review 73, 257-76. Board of Governors of the Federal Reserve System (1943), Banking and Monetary Statistics, 1914-1941. Washington, DC. Bordo, M. D., Erceg, C. J., and Evans, C. L. (2000), “Money, Sticky Wages, and the Great Depression,” American Economic Review 90, 1447-63. Burns, A. F., and Mitchell, W. C. (1946), Measuring Business Cycles. New York: National Bureau of Economic Research. Cecchetti, S. G., and Karras, G. (1994), “Sources of Output Fluctuations During the Interwar Period: Further Evidence on the Causes of the Great Depression,” Review of Economics and Statistics 76, 80-102. Davis, E. P., and Fagan, G. (1997), “Are Financial Spreads Useful Indicators of Future Inflation and Output Growth in EU Countries?” Journal of Applied Econometrics 12, 701-14. Eichengreen, B. (1992), Golden Fetters: The Gold Standard and the Great Depression. New York: Oxford University Press. Eichengreen, B. and Sachs, J. (1985), “Exchange Rates and Economic Recovery in the 1930s,” Journal of Economic History 45, 925-46. Estrella, A., and Hardouvelis, G. A. (1991), “The Term Structure as a Predictor of Real Economic Activity,” Journal of Finance 46, 555-76. Estrella, A., and Mishkin, F. S. (1997), “The Predictive Power of the Term Structure of Interest Rates in Europe and the United States: Implications for the European Central Bank,” European Economic Review 41, 1375-1401. Estrella, A., and Mishkin, F. S. (1998) “Predicting U.S. Recessions: Financial Variables as Leading Indicators,” Review of Economics and Statistics 80, 45-61. Friedman, M., and Schwartz, A. J. (1963), A Monetary History of the United States, 1867-1960. Princeton: Princeton University Press. Friedman, M., and Schwartz, A. J. (1982), Monetary Trends in the United States and the United Kingdom. Chicago: University of Chicago Press.

19

Hamilton, J. D. (1987) “Monetary Factors in the Great Depression,” Journal of Monetary Economics 19, 145-170. Hamilton, J. D., and Kim, D. H. (2002), “A Reexamination of the Predictability of Economic Activity Using the Yield Spread,” Journal of Money, Credit, and Banking 34, 340-360. Harvey, C. R. (1988), “The Real Term Structure and Consumption Growth,” Journal of Financial Economics 22, 305-33. Mishkin, F. S. (1991), “Asymmetric Information and Financial Crises: A Historical Perspective,” in Hubbard, R. G., ed., Financial Markets and Financial Crises. Chicago: University of Chicago Press, 69-108. Morgenstern, O. (1959), International Financial Transactions and Business Cycles. Princeton: Princeton University Press. Plosser, C. I., and Rouwenhorst, K. G. (1994), “International Term Structures and Real Economic Growth,” Journal of Monetary Economics 33, 133-55. Romer, C. D. (1990), “The Great Crash and the Onset of the Great Depression,” Quarterly Journal of Economics 105, 597-624. Stock, J. D., and Watson, M. (1993) “A Procedure for Predicting Recessions with Leading Indicators: Econometric Issues and recent Performance,” in James Stock and Mark Watson (eds.), Business Cycles, Indicators, and Forecasting. Chicago: University of Chicago Press.

20